Skip to content

Latest commit

 

History

History
64 lines (42 loc) · 1.16 KB

README.md

File metadata and controls

64 lines (42 loc) · 1.16 KB

WorldModels

Implementation of World Models paper

Data generation

Initially we collect rollouts of the environment using a random policy.

python datasets/carracing.py

Generated rollouts will be placed in random directory in data/carracing folder

Training

Later we train all the three modules i.e., VAE module, Memory module and Controller module independently

Training the VAE module

python src/train_convVAE.py

Training the Memory module

python src/train_lstm.py

Training the Controller module

python src/evolution_pooling.py

Testing

To test average rewards in gym environment

python src/test_gym.py

Visualisation

To visualise vae original and reconstructed images for analysis

python utils/visualize_vae.py

To plot the graphs from the controller training for analysis

Run utils/plot_utility.ipynb file

Logging

Logging from model training are available in logs folder

Saved Checkpoints

Saved model checkpoints are available in checkpoints folder

References

We have referred ctallec repo